1
$\begingroup$

I would like to know is there any possible approach to recognise digits of the first (and second as well, as you can see in the example figure) order of magnitude. I've thought of neural network with full, half and possibly quarter training samples of the digits I want to recognise, but I don't think this is the best approach. You see, I am new to computer vision and I'd like to see what path should I take to study if there are any for this problem.

Indicator camera shot

$\endgroup$
1
$\begingroup$

If you are using an out-of-the box OCR engine, then there's not much you can do with significantly occluded letters.

However, if you are rolling your own OCR, say with some template-matching, then, there is nothing preventing you from learning/teaching partial letters and their valid combinations (like the 8-9 and 9-0 above). Just treat it as another letter in the set to be recognized.

$\endgroup$
2
  • $\begingroup$ So you recommend using discrete detection of "8,5" and "9,5"? OK. And what about detecting e.g. "8,25" or "8,75". Will some overlapping influence the overall neural network recognition quality? $\endgroup$ – Stat1c_V01D Jun 29 '14 at 19:42
  • $\begingroup$ Actually I was suggesting learning the combination of 2 digits as a single character. Hopefully, smaller conclusions than 1half would be within the generalization capabilities of your detector (I never said NN). $\endgroup$ – Adi Shavit Jun 29 '14 at 19:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.